In a surprising turn of events, Amazon employees are reportedly using an internal AI tool in ways that go beyond the intended purpose. The company introduced a new internal product called MeshClaw, designed to help staff automate routine tasks and improve workflow efficiency. But instead of streamlining work, many are using it to generate unnecessary activity just to boost their AI usage numbers.

Here’s what’s happening. Amazon set a goal for over 80% of its developers to use AI tools weekly. To track progress, they created internal leaderboards based on token consumption—the basic units of data that AI models process. At first, this was meant to encourage adoption, but it’s now backfiring. Employees are firing up MeshClaw to perform trivial or redundant tasks, just to rack up their token counts. This practice, dubbed “tokenmaxxing,” is a classic case of gaming the system.

The Incentive Trap

While Amazon insists that token usage won’t be used in performance reviews, many staff believe management is watching these metrics closely. The pressure to hit usage targets has created a competitive environment. Some workers have confessed to automating activities like email triage or code deployments that don’t genuinely need automation—just to inflate their numbers. It’s a way to look good on paper, even if it doesn’t improve actual productivity.

This situation highlights a common problem when metrics become targets. Instead of motivating real efficiency, they sometimes lead to busywork. When people are rewarded for activity rather than results, they find ways to meet the numbers without adding value. It’s a classic example of what’s called Goodhart’s Law: once a measure becomes a goal, it ceases to be a good measure.

The Broader Implications

Amazon’s case isn’t unique. Similar behaviors are popping up across the tech industry. At other companies, employees are using AI to boost activity metrics, sometimes just to satisfy internal expectations or to outdo colleagues. This trend raises questions about how we measure productivity in the age of AI.

It also sparks a debate about responsible AI use. If teams are incentivized to inflate AI activity, the real benefits of automation—saving time and reducing workload—may never materialize. Instead, AI becomes another checkbox, a way to appear innovative without truly changing how work gets done.

For now, Amazon says it’s committed to the “safe, secure, and responsible” deployment of AI. But the ongoing tokenmaxxing saga shows that when metrics are misaligned, human nature tends to find loopholes. As AI tools become more integrated into daily work, companies will need to rethink how they measure success and ensure these systems serve genuine productivity—not just inflated numbers.